Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!sdd.hp.com!elroy.jpl.nasa.gov!decwrl!fernwood!uupsi!sunic!news.funet.fi!hydra!cc.helsinki.fi!torvalds
From: torvalds@cc.helsinki.fi
Newsgroups: comp.sys.ibm.pc.hardware
Subject: Re: Question about different 386's ...
Message-ID: <1991Apr6.004456.5858@cc.helsinki.fi>
Date: 6 Apr 91 00:44:56 GMT
References: <1991Apr3.180507.22930@bwdls61.bnr.ca> <27fb0c94-b35.1comp.sys.ibm.pc.hardware-1@vpnet.chi.il.us> <1991Apr5.023634.549@agate.berkeley.edu>
Organization: University of Helsinki
Lines: 25

In article <1991Apr5.023634.549@agate.berkeley.edu>, 
c60b-1eq@e260-1f.berkeley.edu (Noam Mendelson) writes:
>> [ discussion on 386DX vs 386SX deleted ]
> 
> I disagree.  Even though the DX has 16 more pins, most of the time they're
> not being used.  This is due to the fact that most MSDOS programs are 16-bit
> (with the exception of 386-specific programs).  So unless you actually make
> use of all 32 pins, a 386SX is not a bad way to go.

I'm not disagreeing outright, but I do have one point: dos (and all
16-bit programs) won't actually use the 16 high bits for data transfer
(the pins are used though), but instruction prefetching goes on as
before. I don't have statistics or anything, but I's say MUCH more
than 50% of all memory fetches are instruction prefetching. Getting 4
bytes instead of 2 at a time into the prefetch queue should make a
notable difference. Anybody have any actual timing differences? The
point is that the 16 extra pins should be a big win even if not used
explicitly by the programmer.

		Linus Torvalds		torvalds@cc.helsinki.fi

PS. Speed shouldn't really be an issue here. As far as I know the SX is
made in max 20MHz versions, most (?) DX's are >20MHz. The big difference
between the DX and SX is in price - if you can live with a slower
system, by all means go for the SX.